Renal biopsy decision-making model construction method for patients with diabetes mellitus complicated with nephropathy
By constructing a kidney biopsy decision-making model based on machine learning algorithms, using multi-center, large-sample real-world data, the representative and extrapolated problems of kidney biopsy decision-making in patients with type 2 diabetes and chronic kidney disease were solved, and effective renal biopsy decision-making and long-term clinical prognosis evaluation were achieved.
Patent Information
- Application Number
- CN202411770408.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-04
- Publication Date
- 2025-03-28
AI Technical Summary
The lack of effective renal biopsy decision-making models in patients with type 2 diabetes and chronic kidney disease leads to underrepresentation of populations and poor practical application extrapolation, and lack of multi-center, large-sample real-world studies to evaluate the long-term clinical prognosis benefits of renal biopsy.
By constructing a kidney biopsy decision-making model based on machine learning algorithms, using multi-center, large-sample real-world hospitalized patient data, a non-diabetic nephropathy risk prediction model is constructed, and a non-diabetic nephropathy risk threshold is proposed to guide clinicians to make kidney biopsy decisions.
Effective renal biopsy decisions for patients with type 2 diabetes and chronic kidney disease have been achieved, which has improved population representation and practical application extrapolation, and provided multi-center, large-sample real-world studies to evaluate the long-term clinical prognosis benefits of renal biopsy, helping to identify patients with non-diabetic nephropathy in the early stage and treat primary diseases.
Smart Images
Figure CN119724577A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of biomedical technology, and in particular relates to a method for constructing a renal biopsy decision model for patients with diabetes combined with nephropathy. Background Art
[0002] Non-diabetic renal disease (NDRD) is not uncommon in patients with type 2 diabetes mellitus (T2DM) and chronic kidney disease (CKD), and its treatment options and clinical prognosis are significantly different from those of diabetic nephrology (DN). Percutaneous renal biopsy is the gold standard for distinguishing NDRD from DN, but there is currently no consensus on the indications for renal biopsy in patients with T2DM and CKD. Many studies have been devoted to building NDRD risk prediction models in patients with T2DM and CKD, in order to assist clinicians in identifying high-risk groups for NDRD and guide renal biopsy decisions. However, the development of these models is limited to special populations and traditional algorithms, and there are deficiencies in population representativeness and practical application extrapolation. In addition, there are currently no multicenter, large-sample, real-world studies at home and abroad to evaluate the long-term clinical prognostic benefits of renal biopsy in patients with T2DM and CKD. Summary of the invention
[0003] In view of this, the present invention aims to overcome the shortcomings of the above-mentioned problems in the prior art, and proposes a method for constructing a renal biopsy decision model for patients with diabetes and nephropathy. The present invention uses multi-center, large-sample real-world inpatient data to guide clinicians in making renal biopsy decisions for patients with type 2 diabetes and chronic kidney disease. To address this clinical need, the present invention first constructs, evaluates and verifies a non-diabetic nephropathy risk prediction model based on a machine learning algorithm in people with type 2 diabetes and chronic kidney disease who undergo renal biopsy, and proposes a non-diabetic nephropathy risk threshold, based on which the overall type 2 diabetes and chronic kidney disease population is stratified for non-diabetic nephropathy risk.
[0004] To achieve the above object, the technical solution of the present invention is achieved as follows:
[0005] The first aspect of the present invention provides a method for constructing a renal biopsy decision model for patients with diabetes and nephropathy, comprising the following steps:
[0006] Step 1: Construct a renal biopsy dataset and a type 2 diabetes combined with chronic kidney disease dataset;
[0007] Step 2: Obtain all the measurement values of the same laboratory test of the same patient within the baseline time window from the data set, and further generate the maximum value, minimum value, median value and the value closest to the index date for each test indicator;
[0008] Step 3: Perform univariate logistic regression on the maximum, minimum, median and most recent value from the index date of each laboratory test indicator, and select the most recent value as the most recent candidate variable;
[0009] Step 4: In the data set of type 2 diabetes and chronic kidney disease, whether or not a renal biopsy was performed was used as a predictor variable. A multivariate logistic regression was constructed with the variables of presence or absence of diabetic retinopathy, rapid decline in eGFR, rapid increase in proteinuria, massive proteinuria, refractory hypertension, hematuria, refractory hypertension, eGFR, and duration of diabetes. The propensity score overlap weighting for each patient to undergo a renal biopsy was calculated. The weight W for the population that had undergone a renal biopsy was 1-PS, while the weight W for the population that had not undergone a renal biopsy was PS.
[0010] Step 5: Divide the renal biopsy dataset into a training set and a validation set, build a renal biopsy decision model in the training set based on weight parameters and the XGBoost algorithm, and verify it in the validation set.
[0011] Furthermore, step 2 also includes interpolation of missing variables. For variables with a missing value ratio of ≤25%, the random forest method is used for interpolation. For variables with a missing ratio of >25%, if it is a categorical variable, the missing part is coded as "unknown"; if it is a continuous variable, it is converted into a corresponding categorical variable based on the clinical reference value or the actual distribution of the data, and the missing part is coded as "unknown".
[0012] Furthermore, in step 5, the renal biopsy data set is divided into a training set (70%) and a validation set (30%) by random stratified sampling according to whether the predictor variable is non-diabetic nephropathy.
[0013] Furthermore, it also includes determining the risk stratification threshold, including:
[0014] According to the prediction value of the renal biopsy decision model, patients with type 2 diabetes and chronic kidney disease were divided into low, medium and high risk groups of non-diabetic nephropathy. The threshold setting of risk stratification comprehensively considered the positive predictive value PPV, negative predictive value NPV, sensitivity SE and specificity SP of the clinical model of multivariate logistic regression in the training set, and 4 groups of threshold combinations were pre-set: threshold combination 1 (NPV 85% + PPV 85%), threshold combination 2 (NPV 90% + PPV 90%), threshold combination 3 (SE 85% + SP 85%) and threshold combination 4 (SE 90% + SP 90%).
[0015] In the data set of the renal biopsy decision model, confusion matrices of four threshold combinations were constructed in turn, and the prediction efficiency of non-diabetic nephropathy was determined by combining different threshold combinations to finally determine the optimal group; if the risk prediction value of non-diabetic nephropathy was less than the low-risk threshold, the patient was classified as the low-risk group of non-diabetic nephropathy and considered to be diagnosed with isolated diabetic nephropathy; if the risk prediction value of non-diabetic nephropathy was higher than the high-risk threshold, the patient was classified as the high-risk group of non-diabetic nephropathy and considered to be diagnosed with non-diabetic nephropathy.
[0016] Furthermore, it also includes applying the renal biopsy decision model to the overall population of people with type 2 diabetes and chronic kidney disease, and stratifying patients for non-diabetic kidney disease risk based on preset risk thresholds.
[0017] Furthermore, it also includes combining the SHAP value importance ranking and adopting the backward selection method to further generate a renal biopsy decision model.
[0018] The second aspect of the present invention provides a device for constructing a renal biopsy decision model for patients with diabetes and nephropathy, comprising:
[0019] Data acquisition module, used to construct renal biopsy dataset and type 2 diabetes combined with chronic kidney disease dataset;
[0020] A data processing module, used for obtaining all measurement values of the same laboratory test of the same patient within the baseline time window from the data set, and further generating the maximum value, minimum value, median value and the most recent value from the index date of each test indicator;
[0021] The variable screening module is used to perform univariate logistic regression on the maximum value, minimum value, median value and the value closest to the index date of each laboratory test indicator, and select the most recent value as the most recent candidate variable;
[0022] The weight determination module is used to use whether or not a renal biopsy was performed as a predictor variable in the data set of type 2 diabetes and chronic kidney disease, to construct a multivariate logistic regression of the variables of diabetic retinopathy, rapid decline in eGFR, rapid increase in proteinuria, massive proteinuria, refractory hypertension, hematuria, refractory hypertension, eGFR, and duration of diabetes, and to calculate the overlapping weighted propensity score for each patient to undergo a renal biopsy. The weight W for the population that has undergone a renal biopsy is 1-PS, while the weight W for the population that has not undergone a renal biopsy is PS.
[0023] The model building module is used to divide the renal biopsy data set into a training set and a validation set, build a renal biopsy decision model in the training set based on weight parameters and the XGBoost algorithm, and verify it in the validation set.
[0024] A third aspect of the present invention provides an electronic device, comprising a processor and a memory that is communicatively connected to the processor and is used to store instructions executable by the processor, wherein the processor is used to execute the above-mentioned method for constructing a renal biopsy decision model for patients with diabetes and nephropathy.
[0025] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned method for constructing a renal biopsy decision model for patients with diabetes and nephropathy.
[0026] Compared with the prior art, the method for constructing a renal biopsy decision model for patients with diabetes and nephropathy described in the present invention has the following advantages:
[0027] The renal biopsy decision model constructed based on the XGBoost algorithm in the present invention and the corresponding non-diabetic nephropathy risk stratification strategy showed good performance, exploring the possible path of machine learning for the scientific management of patients with type 2 diabetes and CKD in the future; in the high-risk group of non-diabetic nephropathy, the implementation of renal biopsy was correlated with better clinical prognosis, further suggesting that the model of the present invention has the potential to become an auxiliary decision-making tool for renal biopsy in patients with type 2 diabetes and CKD, thereby avoiding isolated diabetic nephropathy patients from undergoing unnecessary invasive examinations, assisting in the early identification of non-diabetic nephropathy patients, and promoting non-diabetic nephropathy patients to receive treatment for the primary disease, which is helpful to achieve the goals of hierarchical management and precise diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 The present invention is a flowchart of a method for constructing a renal biopsy decision model for patients with diabetes and nephropathy. DETAILED DESCRIPTION
[0029] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.
[0030] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.
[0031] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.
[0032] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.
[0033] Embodiment 1:
[0034] like Figure 1 As shown, a method for constructing a renal biopsy decision model for patients with diabetes and nephropathy includes the following steps:
[0035] Step 1: Construct a renal biopsy dataset and a type 2 diabetes combined with chronic kidney disease dataset;
[0036] Step 2: Obtain all the measurement values of the same laboratory test of the same patient within the baseline time window from the data set, and further generate the maximum value, minimum value, median value and the value closest to the index date for each test indicator;
[0037] Step 3: Perform univariate logistic regression on the maximum, minimum, median and most recent value from the index date of each laboratory test indicator, and select the most recent value as the most recent candidate variable;
[0038] Step 4: In the data set of type 2 diabetes and chronic kidney disease, whether or not a renal biopsy was performed was used as a predictor variable. A multivariate logistic regression was constructed with the variables of presence or absence of diabetic retinopathy, rapid decline in eGFR, rapid increase in proteinuria, massive proteinuria, refractory hypertension, hematuria, refractory hypertension, eGFR, and duration of diabetes. The propensity score overlap weighting for each patient to undergo a renal biopsy was calculated. The weight W for the population that had undergone a renal biopsy was 1-PS, while the weight W for the population that had not undergone a renal biopsy was PS.
[0039] Step 5: Divide the renal biopsy dataset into a training set and a validation set, build a renal biopsy decision model based on the weight parameter XGBoost algorithm in the training set, and verify it in the validation set.
[0040] In this invention, researchers sourced from the China Renal Data System (CRDS, http: / / www.crds-network.org.cn / # / database) database. CRDS was jointly built by the Nanfang Hospital of Southern Medical University (National Clinical Research Center for Kidney Disease) and the Chronic Disease Center of the Chinese Center for Disease Control and Prevention. As of December 2022, CRDS has covered 25 medical centers in most parts of China, covering the medical treatment data of nearly 9 million inpatients from 2000 to 2022.
[0041] The research population of the present invention was screened from the CRDS database according to strict inclusion and exclusion criteria based on the research purpose.
[0042] T2DM combined with CKD data set (total analysis set): used to describe the baseline characteristics of the T2DM combined with CKD patient population and calculate the weight (Weighting, W) of undergoing renal biopsy.
[0043] Renal biopsy dataset: used to build and validate the renal biopsy decision model.
[0044] Definition of index date: For the renal biopsy population, we defined the time of the first renal biopsy as the index date; for the non-renal biopsy population, we randomly selected the discharge time of any visit to the nephrology or endocrinology department after the diagnosis of T2DM combined with CKD as the index date.
[0045] The present invention defines the 365 days before the index date as the baseline period (for baseline eGFR and urine protein data, we further strictly limit the time window to within 30 days before the index date).
[0046] The data used in the present invention are all from the inpatient and outpatient data systems of each collaborative medical center in CRDS. The database mainly covers the patient's basic demographic information, medical order information, diagnosis information, vital signs information, surgical operation information, electronic medical record text information, laboratory test information, examination information, pathological data and other data.
[0047] In the present invention, the candidate variables of the model mainly involve demographic information, vital signs, diabetes-related information, renal function-related information, clinical comorbidities, combined medications, laboratory blood and urine tests. In addition, all the measured values of the same laboratory test of the same patient in the baseline time window are retained, and the maximum value, minimum value, median value and the value closest to the index date of each test index are further generated.
[0048] Before constructing the renal biopsy decision model, variable pre-screening was performed, and one of the maximum, minimum, median, or most recent values of each laboratory variable was selected as the final candidate variable. Univariate logistic regression was performed on the four levels of each laboratory test index, and the most recent value was selected as the best candidate variable by default. If the remaining three levels of indicators were significantly better than the default value, that is, the difference in Akaike's Information Criterion (AIC) was greater than 10, the one with the smallest AIC was selected as the best candidate variable, and finally a data set for constructing the NDRD risk prediction model was generated.
[0049] The present invention adopts different interpolation methods based on the difference in missing ratio of variables. For variables with missing value ratio ≤ 25%, the random forest method in the R package (Randomforest) is used for interpolation. For variables with missing ratio > 25%, since direct interpolation may cause data deviation, the following processing method is adopted: if it is a categorical variable, the missing part is encoded as "unknown"; if it is a continuous variable, it is converted into a corresponding categorical variable based on the clinical reference value or the actual distribution of the data, and the missing part is encoded as "unknown".
[0050] In the present invention, a candidate variable pre-screening process was performed before the renal biopsy decision model was constructed to select the optimal solution for multiple dimensional data values of each indicator. The final 78 candidate variables and data missing conditions are shown in Table 1.
[0051] Table 1
[0052]
[0053]
[0054]
[0055]
[0056]
[0057] It should be noted that the missing proportion was calculated based on T2DM combined with CKD (sample size 24834). Before building the model, we pre-screened the candidate variables to select one of the maximum, minimum, most recent and median values of each variable as the final candidate variable. Among them, the abbreviations are: BMI, body mass index; eGFR, estimated glomerular filtration rate; ACEI, Angiotensin-Converting Enzyme Inhibitor; ARB, Angiotensin Receptor Blocker.
[0058] In clinical practice, nephrologists usually refer to the 2007 KDOQI guidelines and combine their own clinical experience to perform renal biopsy on patients with T2DM and CKD in a targeted manner. Generally speaking, the renal biopsy population in real-world studies is usually highly selective. Therefore, when constructing an NDRD prediction model in the renal biopsy population, the important factor of the tendency of renal biopsy must be considered in advance, otherwise it is unreasonable to directly apply the prediction model to the T2DM and CKD population in the clinical medical scenario. Based on this, the renal biopsy weight calculation proposed in the present invention mainly refers to the KDOQI guidelines and clinical experience. The specific approach is as follows: In the T2DM combined with CKD data set, whether or not to undergo renal biopsy was used as a predictor variable, and a multivariate logistic regression of the following 9 variables (presence or absence of diabetic retinopathy, rapid decline in eGFR, rapid increase in proteinuria, massive proteinuria, refractory hypertension, hematuria, refractory hypertension, eGFR, and duration of diabetes) was first constructed. Then, the propensity score (PS) for renal biopsy was calculated for each patient. Finally, for those who had undergone renal biopsy, the weight W was 1-PS, while the weight W for those who had not undergone renal biopsy was PS.
[0059] In the present invention, referring to the KDOQI guidelines and clinical prior knowledge, a multivariate logistic regression including the 9 variables mentioned above was constructed in the T2DM combined with CKD data set to obtain the propensity score for each patient to undergo renal biopsy, and the weight of each patient to undergo renal biopsy was further calculated based on this score.
[0060] The present invention introduces a renal biopsy weight parameter W when constructing a renal biopsy decision model to correct the NDRD risk probability value.
[0061] When constructing the model, the present invention adopts a random stratified sampling method to divide the renal biopsy data set into a training set accounting for 70% and a validation set accounting for 30% according to the predictor variable (whether it is NDRD).
[0062] The present invention pre-constructed an XGBoot full model with a total of 78 variables. According to the SHAP value of each feature, the first 40 features were selected to further explore the correlation between the number of feature values and the model prediction efficiency based on the backward selection method. Finally, 13 feature numbers were selected as predictors for constructing the XGBoost parsimonious model, including diabetes course, hemoglobin, diabetic retinopathy, insulin use, serum creatinine, serum albumin, urine total protein concentration, urine glucose test strip qualitative detection, urine occult blood test strip qualitative detection, age, systolic blood pressure, serum IgG and serum IgA. The renal biopsy decision model was constructed in the training set using population weights and based on the XGBoost algorithm combined with 13 predictors, and verified in the validation set.
[0063] Embodiment 2:
[0064] A device for constructing a renal biopsy decision model for patients with diabetes and nephropathy, including:
[0065] Data acquisition module, used to construct renal biopsy dataset and type 2 diabetes combined with chronic kidney disease dataset;
[0066] A data processing module, used for obtaining all measurement values of the same laboratory test of the same patient within the baseline time window from the data set, and further generating the maximum value, minimum value, median value and the most recent value from the index date of each test indicator;
[0067] The variable screening module is used to perform univariate logistic regression on the maximum value, minimum value, median value and the value closest to the index date of each laboratory test indicator, and select the most recent value as the most recent candidate variable;
[0068] The weight determination module is used to use whether or not a renal biopsy was performed as a predictor variable in the data set of type 2 diabetes combined with chronic kidney disease, to construct a multivariate logistic regression of the variables of presence or absence of diabetic retinopathy, presence or absence of rapid decline in eGFR, presence or absence of rapid increase in proteinuria, presence or absence of massive proteinuria, presence or absence of refractory hypertension, presence or absence of hematuria, presence or absence of refractory hypertension, eGFR, and duration of diabetes, and to calculate the propensity score for each patient to undergo a renal biopsy. The weight W for the population that has undergone a renal biopsy is 1-PS, while the weight W for the population that has not undergone a renal biopsy is PS;
[0069] The model building module is used to divide the renal biopsy data set into a training set and a validation set, build a renal biopsy decision model in the training set using population weights and based on the XGBoost algorithm, and verify it in the validation set.
[0070] Embodiment three:
[0071] An electronic device comprises a processor and a memory which is in communication with the processor and is used to store instructions executable by the processor. The processor is used to execute the above-mentioned method for constructing a renal biopsy decision model for patients with diabetes and nephropathy.
[0072] Embodiment 4:
[0073] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned method for constructing a renal biopsy decision model for patients with diabetes and nephropathy.
[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for constructing a renal biopsy decision model for patients with diabetes and nephropathy, characterized in that: The steps include: Step 1: Construct a renal biopsy dataset and a type 2 diabetes combined with chronic kidney disease dataset; Step 2: Obtain all the measurement values of the same laboratory test of the same patient within the baseline time window from the data set, and further generate the maximum value, minimum value, median value and the value closest to the index date for each test indicator; Step 3: Perform univariate logistic regression on the maximum, minimum, median and most recent value from the index date of each laboratory test indicator, and select the most recent value as the most recent candidate variable; Step 4: In the data set of type 2 diabetes and chronic kidney disease, whether or not a renal biopsy was performed was used as a predictor variable. A multivariate logistic regression was constructed with the variables of presence or absence of diabetic retinopathy, rapid decline in eGFR, rapid increase in proteinuria, massive proteinuria, refractory hypertension, hematuria, refractory hypertension, eGFR, and duration of diabetes. The propensity score overlap weighting for each patient to undergo a renal biopsy was calculated. The weight W for the population that had undergone a renal biopsy was 1-PS, while the weight W for the population that had not undergone a renal biopsy was PS. Step 5: Divide the renal biopsy dataset into a training set and a validation set, build a renal biopsy decision model in the training set based on weight parameters and the XGBoost algorithm, and verify it in the validation set.
2. The method for constructing a renal biopsy decision model for patients with diabetes and nephropathy according to claim 1, characterized in that: The step 2 also includes interpolation of missing variables. For variables with a missing value ratio of ≤25%, the random forest method is used for interpolation. For variables with a missing ratio of >25%, if they are categorical variables, the missing part is coded as "unknown"; if they are continuous variables, they are converted into corresponding categorical variables based on clinical reference values or actual data distribution, and the missing part is coded as "unknown".
3. The method for constructing a renal biopsy decision model for patients with diabetes and nephropathy according to claim 1, characterized in that: In step 5, the renal biopsy data set is divided into a training set (70%) and a validation set (30%) by random stratified sampling according to whether the predictor variable is non-diabetic nephropathy.
4. The method for constructing a renal biopsy decision model for patients with diabetes and nephropathy according to claim 1, characterized in that: It also includes determining risk stratification thresholds, including: According to the prediction value of the renal biopsy decision model, patients with type 2 diabetes and chronic kidney disease were divided into low, medium and high risk groups of non-diabetic nephropathy. The threshold setting of risk stratification comprehensively considered the positive predictive value PPV, negative predictive value NPV, sensitivity SE and specificity SP of the clinical model of multivariate logistic regression in the training set, and 4 groups of threshold combinations were pre-set: threshold combination 1 (NPV 85% + PPV 85%), threshold combination 2 (NPV 90% + PPV 90%), threshold combination 3 (SE 85% + SP 85%) and threshold combination 4 (SE 90% + SP 90%). In the data set of the renal biopsy decision model, confusion matrices of four threshold combinations were constructed in turn, and the prediction efficiency of non-diabetic nephropathy was determined by combining different threshold combinations to finally determine the optimal group; if the risk prediction value of non-diabetic nephropathy was less than the low-risk threshold, the patient was classified as the low-risk group of non-diabetic nephropathy and considered to be diagnosed with isolated diabetic nephropathy; if the risk prediction value of non-diabetic nephropathy was higher than the high-risk threshold, the patient was classified as the high-risk group of non-diabetic nephropathy and considered to be diagnosed with non-diabetic nephropathy.
5. The method for constructing a renal biopsy decision model for patients with diabetes and nephropathy according to claim 1, characterized in that: It also includes applying the renal biopsy decision model to the overall population of type 2 diabetes and chronic kidney disease, and stratifying patients for non-diabetic kidney disease risk based on preset risk thresholds.
6. The method for constructing a renal biopsy decision model for patients with diabetes and nephropathy according to claim 1, characterized in that: It also includes combining SHAP value importance ranking and taking backward selection method to further generate a renal biopsy decision model.
7. A device for constructing a renal biopsy decision model for patients with diabetes and nephropathy, characterized in that: include: Data acquisition module, used to construct renal biopsy dataset and type 2 diabetes combined with chronic kidney disease dataset; A data processing module, used for obtaining all measurement values of the same laboratory test of the same patient within the baseline time window from the data set, and further generating the maximum value, minimum value, median value and the most recent value from the index date of each test indicator; The variable screening module is used to perform univariate logistic regression on the maximum value, minimum value, median value and the value closest to the index date of each laboratory test indicator, and select the most recent value as the most recent candidate variable; A weight determination module is used to use whether or not a renal biopsy is performed as a predictor variable in a data set of type 2 diabetes and chronic kidney disease, to construct a multivariate logistic regression with the variables of presence or absence of diabetic retinopathy, rapid decline in eGFR, rapid increase in proteinuria, massive proteinuria, refractory hypertension, hematuria, refractory hypertension, eGFR, and duration of diabetes, to calculate the overlapping weighted propensity score for each patient to undergo a renal biopsy, with the weight W for those who have undergone a renal biopsy being 1-PS, and the weight W for those who have not undergone a renal biopsy being PS; The model building module is used to divide the renal biopsy data set into a training set and a validation set, build a renal biopsy decision model in the training set based on weight parameters and the XGBoost algorithm, and verify it in the validation set.
8. An electronic device, comprising a processor and a memory connected to the processor for storing instructions executable by the processor, characterized in that: The processor is used to execute the method for constructing a renal biopsy decision model for patients with diabetes and nephropathy as described in any one of claims 1 to 6.
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing a renal biopsy decision model for patients with diabetes and nephropathy described in any one of claims 1 to 6 is implemented.